Pension Entitlements and Net Wealth in

(1)

WORKING PAPER 238

OESTERREICHISCHE NATIONALBANK

E U R O S Y S T E M

Pension Entitlements and Net Wealth in

Austria

(2)

The Working Paper series of the Oesterreichische Nationalbank is designed to disseminate and to provide a platform for discussion of either work of the staff of the OeNB economists or outside contributors on topics which are of special interest to the OeNB. To ensure the high quality of their content, the contributions are subjected to an international refereeing process. The opinions are strictly those of the authors and do in no way commit the OeNB.

The Working Papers are also available on our website (http://www.oenb.at) and they are indexed in RePEc (http://repec.org/).

Publisher and editor Oesterreichische Nationalbank

Otto-Wagner-Platz 3, 1090 Vienna, Austria PO Box 61, 1011 Vienna, Austria

www.oenb.at oenb.info@oenb.at

Phone (+43-1) 40420-6666 Fax (+43-1) 40420-046698

Editor Martin Summer

Cover Design Information Management and Services Division

DVR 0031577

ISSN 2310-5321 (Print) ISSN 2310-533X (Online)

(3)

Pension Entitlements and Net Wealth in Austria ^∗

Markus Knell

^∗∗

and Reinhard Koman

^∗∗∗

Oesterreichische Nationalbank February 2022

Abstract

This study combines data from the HFCS (Household Finance and Consumption Survey) and the social security registry to estimate the present value of public pension entitlements for Austria in the year 2017. The household averages of the present value of pension entitlements and of private net wealth turn out to be similar (both amounting to arounde250,000) which is in line with the results for other countries like Switzerland, Germany and the US. Since pension entitlements are more equally distributed than other assets most inequality measures for augmented wealth (the sum of pension entitlements and net wealth) are lower than for net wealth. The Gini coefficient for Austria, e.g., decreases from 0.73 (for net wealth) to 0.53 (for augmented wealth) which is again fairly similar to the results for other countries. Furthermore, it is shown that the main results are robust to many alternative specifications. In particular, estimates based on statistical matching and on direct survey information lead to surprisingly similar results. The same is true for specifications with homogeneous or heterogeneous life expectancy and with retirement at the statutory or the individually expected retirement age. Finally, the paper compares the results to the ones of the related literature, sums up the limitations of the approach and discusses why the results have to be interpreted cautiously due to the fact that pension entitlements and net wealth are not perfectly commensurable concepts.

Keywords: Net wealth, Net worth, Pension entitlements, Augmented wealth, Life cycle, HFCS

JEL-Classification: D31, H55, J32

∗The study is based on a joint project by the Economic Analysis and Research Department of the Oesterreichische Nationalbank. We thank Governor Prof. Robert Holzmann who initiated the project and provided many valuable comments and suggestions. We are also grateful to Ursina Kuhn for a number of detailed and highly useful remarks. The views expressed in the study do not necessarily reflect those of the Oesterreichische Nationalbank.

∗∗OeNB, Email: [email protected]. Corresponding author.

∗∗∗OeNB, Email: [email protected].

(4)

Non-Technical Summary

In Austria, as in many other countries, there exists a well-developed public pension system that is the main source of old-age income for a majority of the population. Both theoretical models and basic intuition suggest that the presence of a well-developed and credible pension system will decrease the necessity to accumulate private wealth which should be reflected in individuals’ decisions on life-cycle savings. In recent years a number of papers have tried to quantify the aggregate value of these public pension promises and to contrast them to the existing estimates of net wealth (i.e. the sum of financial and real asset minus total debts). So far this has only been done for a small group of countries including Germany, Switzerland, Italy, the US and Australia. In this present paper we add to this literature by conducting a similar exercise for Austria. To this end we combine data from the Household Finance and Consumption Survey for the year 2017 with data from the social security register.

The paper has three main contributions. First, we complement the existing literature on augmented wealth (i.e. the sum of pension entitlements and private net wealth) by expanding it to yet another country. We find that public pension entitlements are important for the average Austrian household. In our benchmark specification we estimate their value to being of almost the same size as private net wealth (both amounting to around e250,000). This underlines the size and importance of the public pension system for old-age security in Austria. Since most households receive pensions or have pension claims and since these pension entitlements are more equally distributed than other assets, most inequality measures for augmented wealth are lower than for net wealth. The Gini coefficient for Austria, e.g., decreases from 0.73 (for net wealth) to 0.53 (for augmented wealth).

The second contribution of the paper is the provision of a systematic overview of the existing literature, both with respect to the employed methods (Table 1) and to the estimated results (Table 10). We document that the cross-country results show a similar pattern for the majority of countries. In particular, we find that the share of pension entitlements in augmented wealth typically lies round 50% (as is the case for Austria). One difference across countries is, however, whether the value of pension entitlements is concentrated in the first (the public) pillar (as in Austria) or in the second and third pillars. For the distributional measures the pattern is also quite similar across most countries. The Gini coefficient, e.g., is reduced when moving from net wealth to augmented wealth by around 30% for Austria, Germany and Switzerland.

The third contribution of the paper is methodological. We show that the basic results are quite robust to various assumptions concerning life expectancy, the retirement age and the data source. In particular, the use of statistically matched data (our benchmark approach) leads to almost identical results as the use of direct survey information or an estimation of expected pension benefits based on recollected work history. These robustness results are relevant for cross-country comparisons since the papers of the related literature

(5)

1 Introduction

In Austria, as in many other countries, there exists a well-developed public pension system that is the main source of old-age income for a majority of the population. If (or in as far as) the current stipulations of the system are taken as credible, these pension promises will reduce the necessity and the incentives to engage in life-cycle savings, thereby also reducing the fraction of wealth which is accumulated for this purpose. In recent years a number of papers have tried to quantify the aggregate value of these public pension promises and to contrast them to the existing estimates of net wealth (i.e. the sum of financial and real asset minus total debts).¹ This has been done, e.g., for Germany (Rasner et al. 2013, Bönke et al. 2019, Bönke et al. 2020), Switzerland (Kuhn 2020), the US (Bönke et al. 2020, Sabelhaus & Volz 2020, Catherine et al. 2020) and Australia (Longmuir 2021). In the present paper we add to this literature by conducting a similar exercise for Austria. To this end we combine data from the third wave of the Household Finance and Consumption Survey (the HFCS, which contains information about private wealth for Austrian households from the year 2016/2017) with data from the social security register (which provides information about public pension entitlements for the same period). Since it has not been possible to directly link the survey and the administrative data (e.g. via a unique identifier) we had to resort to the method of statistical matching in order to approach this task.

In line with the results of the related literature we find that public pension entitlements are important for the average Austrian household. In our benchmark estimation we find that their value is of almost the same size as private net wealth (both amounting to arounde250,000). This underlines the size and importance of the public pension system for old-age security in Austria. Our results are completely in line with the ones for Germany and Switzerland that have found a similar quantitative importance of pension entitlements relative to private net wealth. One difference across countries is, however, whether the value of pension entitlements is concentrated in the first (the public) pillar (as in Austria) or in the second and third (the occupational and private) pillars as, e.g., in Switzerland. The presence of pension entitlements also affects the estimation of the distribution of resources across households. If one adds the value of pension entitlements to the value of net wealth then the resulting entity of “augmented wealth”² is more equally

1The literature sometimes uses the term “net worth” instead of net wealth. Acknowledging that there exist different definitions of these expressions we use them synonymously in the following and we will mainly stick to the term “net wealth”.

2This term has been suggested by Wolff (1996) and Davies & Shorrocks (2000) and has been taken up

(8)

distributed than private wealth. In our analysis, e.g., the Gini coefficient for augmented wealth comes out as 0.53 while the Gini coefficient for net wealth is 0.73. These results are again comparable with the ones for Germany (where the Gini is reduced from 0.76 to 0.51) and Switzerland (where it drops from 0.75 to 0.55).

The inclusion of pension entitlements offers are more encompassing picture of households’ economic possibilities across the life cycle. Households that expect sizable future public pension benefits are likely to hold less assets than households that have to provide by themselves for old-age security. The exclusion of public pensions might thus distort the assessment of the distribution of economic resources between households within a country and also the comparison between countries that are characterized by differently organized welfare states (see Fessler & Sch¨urz 2018). At the same time it has to be noted at the outset that the combination of private net wealth and (public) pension entitlements is not without problems and the ensuing results have to interpreted with care. The first concern is of a more practical nature and refers to the difficulty of coming up with reliable estimates of future pension entitlements, especially if the goal is to produce an internationally comparable dataset. This concern is, e.g., emphasized in the Guidelines for Micro Statis- tics on Household Wealth by the OECD (2013).³ The objection against the integration of public pension rights into households goes, however, considerably beyond these practical considerations. These fundamental objections focus on three important issues. The first one is that the valuation of wealth should be based on a concept of “marketable wealth”.

There does not exist a market for future pension entitlements and each calculation must be based on a number of specific and ultimately arbitrary assumptions.⁴ Second, once the analysis is extended beyond the items for which markets and prices exist, it is no longer clear where to draw the line.⁵ The third issue is even more fundamental and refers

frequently, e.g. also by the OECD (2013). The standard definition of augmented wealth adds public and occupational pension entitlements to net wealth. We want to clarify at the outset, however, that for our estimates of augmented wealth in Austria we abstract from occupational pension entitlements. We do this for reasons of data availability. This omission, however, is likely to be innocuous since occupational pensions play only a minor role for Austrian pensioners. We come back to this issue in section 6.4 where we also present estimates based on rudimentary data that are in line with this conjecture.

3“The exclusion of entitlements in social security schemes, as recommended here for micro statistics on household wealth, is primarily for practical reasons and to maintain consistency with the SNA’s [System of National Accounts’] definition of financial assets. It reflects the view that reliable estimates of pension entitlements in social security schemes may not be readily available in many countries.” (OECD 2013, p.71).

4“Once we depart from observed market transactions, any estimate of what assets ‘would sell for’

involves a number of speculative assumptions. This applies to various classes of assets but is particularly the case with defined benefit pension rights” (Alvaredo et al. 2018, p.28).

5“Including Social Security in wealth would thus call for including the present value of future health

(9)

to the different functions, disposabilities and capabilities of various categories of wealth.

Financial assets are not only useful as a store of value for future consumption needs, they are typically also instantaneously available in case of an unforeseen emergency, as a down-payment for the purchase of real estate or the founding of a business enterprise.

Furthermore, they can be inherited to the next generation and they can—depending on their size—also be drawn on to maintain and gain social status and to try to exert influence in the social or political arena.⁶ Future pension entitlements do not fulfill any of these additional functions of wealth and based on this observation they should not get the same weight in a compilation of household wealth.⁷ Ultimately, it depends on the focus of the investigation whether and to which extent pension entitlements should be included into the analysis. All of these obstacles and concerns are well-known and also briefly discussed in some papers of the related literature.⁸ Typically, however, these objections are only mentioned in the introductory remarks while the rest of the papers treats the present value of public pension entitlements (often referred to as “pension wealth”) as completely equivalent to the rest of the items of household wealth. In order to remind us and the reader of the fact that pension promises are a different entity than net wealth we stick in this paper to the term “(public) pension entitlements”. We will use the expression

“augmented wealth” for the sum of pension entitlements and net wealth, however, since it is an established notion (see Wolff 1996, Davies & Shorrocks 2000, OECD 2013). Also in this case, however, one should remain aware of the fact that this magnitude adds up two incommensurable, or at least heterogeneous entities.

The paper is structured as follows. In section 2 we summarize the existing literature and we also delineate the available methods that can be used to incorporate information on households’ pension entitlements. In section 3 we present the basic structure of the Austrian pension system while in section 4 we describe the two main data sources—the

benefits (such as Medicare benefits in the United States), future government education spending for one’s children, etc., net of future taxes. It is not clear where to stop, and such computations are inherently fragile because of the lack of observable market prices for these types of assets” (Zucman 2019, p.113).

6“It has also to be remembered that we are concerned about the distribution of wealth not only on account of the potential consumption. Wealth conveys power. [. . .] The degree of direct personal control over resources [. . .] is one of the major reasons for interest in the concentration of wealth.[. . .] It is [then]

reasonable to omit assets, such as pension rights, over which the individual has only limited or no control”

(Alvaredo et al. 2018, p.28).

7One could argue that the existence of survivor pensions provides for heritability of pension rights.

This, however, would ignore major differences between survivor pensions and normal bequests. The recipients of survivor pensions cannot be freely chosen, the claims cannot be passed down to further generations and the eligibility depends on various conditions (e.g. income differences, remarriage etc.).

8See, for example: B¨onke et al. (2020, p.38), Kuhn (2020, p.1), Catherine et al. (2020, p.3). See also Fessler et al. (2011).

(10)

HFCS and data from the social security registry—and we also sketch the technique of statistical matching used to combine these two data sets (details can be found in Lindner

& Sch¨urz 2021). Section 5 discusses the basic formula that is used to calculate the present value of public pension entitlements. We show how the estimates are based on the choice of several crucial parameters (the retirement age, the survival rates, the discount rate and the definition of pension benefits) and we state (and justify) the assumptions underlying our benchmark estimation. Section 6 contains the results for our estimates for Austria, divided in various sub-sections dealing with: the results for the aggregate values and the implications for the wealth composition (section 6.1), a breakdown with respect to various socio-demographic characteristics (section 6.2), the distribution of public pension entitlements across the population and various inequality measures (section 6.3). In section 6.4 we show how the benchmark results change for various alternative assumptions (with respect to the definition of pension income, life expectancy, the retirement age and the discount rate). We show there that the basic results are quite robust to the use of different data sources and methods to calculate the pension entitlements. in particular, the use of statistically matched data (our benchmark approach) leads to almost identical results as the use of direct survey information or an estimation of expected pension benefits based on survey respondents’ recollected work history. These novel robustness results are relevant for cross-country comparisons since the papers of the related literature are often based on different methods and approaches to calculate pensions entitlements.

We also show that the benchmark results remain almost constant for the assumptions of homogeneous instead of heterogeneous (i.e. income-dependent) life expectancy (as in the benchmark) and the same is true if we use the individually expected retirement age instead of the statutory retirement age (as in the benchmark). In section 6.5 we discuss a number of limitations of our approach (referring, e.g., to the exclusion of minimum and survivor pensions). In section 7 we provide an tabular overview of the related literature and we compare our results to the ones for other countries. This part is the second contribution of the paper that goes beyond the documentation of country-specific estimations for Austria. The international comparison shows a broad similarity of the results for the majority of countries and also offers some tentative explanations for the smaller group of countries that do not follow the general pattern. Section 8 finally concludes and discusses implications of the findings.

(11)

2 Related literature

The literature on the calculation of public pension entitlements is not excessively large and there exist only a small number of papers for a handful of countries. What is more, we are only aware of one paper (B¨onke et al. 2020) which attempts to calculate comprehensive and comparable measures of (public) pension entitlements for more than one country (in this case Germany and the US). This has to do with the fact that both—wealth surveys and even more pension systems—are highly country-specific and it is often quite diffi- cult to make them comparable. Table 1 provides a brief summary of the main papers of the related literature (including—for the sake of comparison—in the last line the present paper).⁹ The table reports the country, the time period, the unit of observation (individual or household), the sample size, the data source and the method used to combine information on net wealth (defined as the sum of financial and real assets minus eventual debts) and on public pension entitlements. All papers use comprehensive national surveys to come up with estimates of net wealth while the methods employed to append public pension entitlements differ along two dimensions: the source of data (the survey itself, linked register data or statistically matched administrative data) and the construction of the pension entitlements (direct information on the present value or simulations based on individual work histories and prevailing regulations). In principle there could thus be six possible combinations of data source and pension calculation, although not all of them can actually be observed in the literature. The choice for one or another method depends mainly on data availability, institutional details and legal restrictions. In the following we provide a brief overview of the most popular methods ordered by the underlying data source.

Information on pension entitlements in wealth surveys: The first and self-evident possibility to amend the traditional wealth data recommends itself when the survey itself includes reliable information that can be used to estimate pension entitlements.

• Survey questions on present values: If the survey contains a specific question on the

9A number of papers have not been included below since they focus on special subgroups of the population: Maunu (2010), e.g., only includes non-retired Finnish households above the age of 45; Crawford

& Hood (2016) British individuals aged between 65 and 79; Wolff (2015) and Jacobs et al. (2021) US households for age brackets between 40 and 64; Cowell et al. (2017) European households (in 13 countries) whose reference person is aged 65-84. Roine & Waldenstr¨om (2009) have a different perspective as they focus on the development of wealth concentration in Sweden over the long-term (1873-2006).

An early attempt to quantify social security wealth for Austria is Holzmann (1981) which—due to data limitations—is based on national income data.

(12)

Table1:Mainfeaturesoftherelatedliterature PaperCountryPeriodUnitSamplesizeDataSourceMethod Mazzaferro&Toso(2009)ITA1991-2002HH8,000SHIWLinkagesurvey/admin.data Rasneretal.(2013)DEU2007Ind.20,000SOEP/admin.dataStat.Matching Bönkeetal.(2019)DEU2012/13Ind.16,200SOEPAdmin.inform.plusimput. Bönkeetal.(2020)DEU2012HH8,500SOEPAdmin.inform.plusimput. Bönkeetal.(2020)USA2012HH6,000SCFAdmin.inform.plusimput. Sabelhaus&Volz(2020)USA1995-2016HH4,000-6,000SCFRetrosp.workhist. Catherineetal.(2020)USA1989-2016HH4,000-6,000SCFRetrosp.workhist. Kuhn(2020)CHE2015Ind./HH10,164/7,468SILC/admin.dataLinkagesurvey/admin.data Longmuir(2021)AUS2018HH9,486HILDAsurveyinformation Thispaper(2022)AUT2017HH3,072HFCS/admin.dataStat.Matching Note:Theabbreviationsforthedatasourcesareasfollows:SHIW=SurveyofHouseholdIncomeandWealth,SOEP=Socio-Economic Panel,SCF=SurveyofConsumerFinances,SILC=StatisticsonIncomeandLivingConditions,HILDA=Household,IncomeandLabour DynamicsinAustralia,HFCS=HouseholdFinanceandConsumptionSurvey.Otherabbreviationsare:HH=households,ind=individuals, admin=administrative,stat=statistical,inform=information,imput=imputation,retrosp=retrospective,hist=history.

(13)

present value of future pensions entitlements then this approach is straightforward.

The available responses can be treated like the other wealth components and sim- ply added to the calculations. The problem with this approach is that it is often not enough to ask respondent about their pension entitlements since many individuals (especially when retirement seems far away in the future) do not have deep knowledge about their accrued pension entitlements (or—depending on the nature of the pension system—the benefits are not even easily observable in advance and are only calculated at the moment of retirement). It would thus be beneficial in this situation if the interviewer could resort to official information (e.g. pension account statements) either via inspection of documents provided by the respondents or via their agreement to access register information. This has, e.g., been done in special waves of the German Socio-Economic Panel (SOEP). B¨onke et al. (2019) report that 41% of the respondents looked at the official information from the Gesetzliche Rentenversicherung to indicate the value. The paper uses imputed values if a respondent did not report a value or only provided an approximate value. This means, however, that the share of imputed pension information might still be fairly large even if respondents are asked (or motivated) to provide the official information.

• Survey questions on retrospective work history: Sometimes surveys include information about respondents’ past labor market experiences (like the start of their working career, their spell of unemployment or non-work periods and their received wages).

This allows the researchers to calculate (or rather simulate) the expected pension benefits by using the existing legislation of the pension system. This approach is rather tedious since it requires not only the careful processing of the information of past work history (and also the filling in of missing information) but also an exact coding of the regulation that typically includes a good number of details, exceptions, special treatments etc. Furthermore, not many surveys include sufficiently detailed information such as to facilitate this approach. One exception is the US Survey of Consumer Finances (SCF) that includes a module that contains the retrospective work history and prospective work expectations of respondents in the SCF.

This method has been chosen, for example, by Sabelhaus & Volz (2020), Catherine et al. (2020) and Jacobs et al. (2021). These authors also use various methods of validation to show that this approach leads to satisfying results.¹⁰

10In particular, the authors state that they “can match the aggregate estimate of Social Security wealth of the SSA [Social Security Administration], and that [their] estimates correctly match actual retirement-age benefits reported in the SCF” (Catherine et al. 2020, p.2).

(14)

Exact link to register data: The quality of the results can typically be improved if it is possible to use precise register data to get the necessary information about pension entitlements. This could, e.g., be done by using a unique identifier (like the social security number) that is present both for the survey respondents and also in the official register data. This approach has been used, e.g., by Kuhn (2020) for Swiss data. In particular, the author linked survey data for Switzerland coming from the SILC survey conducted in 2015 with various types of administrative data (including the registries for federal income, population, marriage, divorce, birth and death). The matching rate was an impressive 99% of the sample.¹¹ In processing the data one can again use the two possible ways to extract pension entitlements.

• Register data on present values: This approach is possible if the linked administrative data already contain present values of pensions entitlements or information that is closely related to these values, like—for example—the total pension points in a point system (cf. Germany), the pension account value in a notional defined contribution system (cf. Sweden) or the total credits in a notional defined benefit system (cf. Austria). In Germany there exists an ongoing project that follows this route by linking data from participants in the SOEP survey to their individual record in the pension insurance (L¨uthen et al. 2022, forthcoming).

• Register data on work history: Even if the register data might not provide direct present values of entitlements they can sometimes still be used to provide more accurate estimates based on work history. This method (that again involves many assumptions and the coding of pension regulations) has, e.g., been followed by Kuhn (2020) in her work with Swiss register data.

Statistical matching: The final method of adding pension entitlements to wealth data is the use of statistical matching techniques. This is the method at hand if the survey neither contains reliable pension data nor sufficient work history information and if the exact linkage of survey and administrative data is impossible for technical or legal reasons.

Again the statistically matched data can involve present values of pension entitlements or data on work history that have to be transformed into entitlement estimations. This method is followed, e.g., by Rasner et al. (2013) who statistically match administrative

11The author states that “two reasons explain the high matching rates for data linkage: social security number was contained in the sampling frame and the linkage required no additional consent from survey respondents” (p.5). In particular, the linkage of the SILC data and the administrative records was based on a project-specific contractual agreement.

(15)

data from the Gesetzliche Rentenversicherung to the German SOEP.¹²

For the Austrian data we also use the last approach based on statistical matching as our main method. The second method—based on a direct link between survey respondents and administrative data—has been impossible due to legal restrictions. The first method—based on survey information—was not chosen as the primary approach for practical reasons. On the one hand, the HFCS for Austria contains a block of question on the acquired public pension claims. The use of this direct survey information, however, proved to be problematic since many respondents did not give an answer to this question and, furthermore, the received answers do not seem to be completely reliable. The construction of pension entitlements based on retrospective work history, on the other hand, did not look promising since the HFCS contains only very sparse information about past labor market variables. In section 6.4 we analyze, however, how the benchmark results based on statistical matching change if we use the alternative methods based on (incomplete) direct survey information or (very rough) measures of the work history. Anticipating the results we find that the different methods lead to surprisingly similar estimates. This is an interesting finding that not only increases the confidence in the benchmark results of this paper but also supports the significance of cross-country comparisons that are based on papers that use different methods.

In closing this literature review we want to briefly mention that there exists an additional strand of literature that is related to our investigation. In particular, this literature focuses on one implication of the life-cycle model and tries to analyze the effect of public pension entitlements on private savings. The benchmark model (Modigliani 1986) suggests a one-to-one substitution between the two magnitudes. The empirical literature—starting with Feldstein (1974)—finds somewhat mixed results depending on the definition of future pension entitlement (in the US context often classed SSW, “Social Security Wealth”), the studied time period and the empirical method used. Some well-known papers in this literature include Attanasio & Brugiavini (2003), Bottazzi et al. (2006) and Chetty et al.

(2014). The literature is too vast to be summarized here. Overall one can say that the studies find only weak evidence for a crowing out effect, typically considerably lower than the full set-off implied by the theoretical model. This is explained by additional savings

12In particular, they used the so-called “Versichertenkontenstichprobe” that contains the data of about 1% of the insured population in Germany. The authors use 4 different matching techniques (hotdesk, regression, predictive mean matching, Mahalanobis distance) and come to the conclusion that the latter measure shows the best performance.

(16)

motives (bequest, precaution etc.), liquidity constraints, risk-aversion, individual myopia, incomplete information and less-than-rational behavior. For the context of our analysis it has to be stressed that these papers are not so much concerned with the effect of future pension entitlements on the estimation of average wealth or the wealth distribution but rather with a test of behavioral predictions in the context of the life-cycle model.

Finally, we want to mention that the period OECD publicationsPensions at a Glance (e.g. OECD 2019) also include numbers for country-specific pension wealth. These, however, are based on hypothetical individuals with stylized employment careers and are thus not directly related to the orientation of this paper.

3 The Austrian pension system

The majority of the working population in Austria participates in a mandatory public pension scheme that is specified in the General Pension Act (Allgemeines Pensionsgesetz, APG). Some (liberal) professions like doctors and lawyers have separate systems that follow specific rules and that are excluded from our calculations. Civil servants also had and in certain areas still have separate system which are briefly described in appendix A.3.2. Occupational pensions only play a minor role in the current Austrian pension landscape. In the year 2016 the total accumulated asset of the second pillar amounted to only 6% of GDP (considerably below the OECD average of 100%) and only 10% of pensioners received any occupational pension benefits. In the following we abstract from these occupational pension benefits but come back to the issue in section 6.4.

The APG system is organized on a pay-as-you-go basis and it is based on “individual defined benefit pension accounts”. The contribution rate stands at 22.8%, of which 10.25%

are paid by employees and 12.55% by employers (there are some exceptions for farmers and for self-employed persons). The main element of the benefit side of the system is the formula: “80/65/45”: After 45 years of insurance and retirement at the age of 65, the system provides an initial pension benefit that corresponds to 80% of average lifetime income. This target is implemented by means of an accrual rate (“Kontoprozentsatz”).

Every year 1.78% of total earnings (up to a ceiling, the “H¨ochstbeitragsgrundlage”)¹³ are credited to the account (“Teilgutschrift”, annual credit AC) while past credits are revalued by the growth rate of the average contribution basis. The revalued past credits and the annual new credit add up to the total pension value (“Gesamtgutschrift”, total credit

13The ceiling in the year 2016 was set ate4,860 per month or e68,040 per year (14 monthly install- ments). About 5% of all employees receive earnings that exceed this upper threshold.

(17)

TC). The credits are recorded in the pension accounts that can be retrieved either via online access or (on demand) by normal mail by the insurance agency. The information given in the pension account statement can be easily transformed into pension benefit levels. In fact, by construction the total credits correspond to the annual initial pension benefit that an insured person could expect if he or she retires at the statutory retirement age (65 years for males and increasing until 2033 from 60 to 65 for women) and if there would not be any additional credits to the account and no further revaluation. Although these total credits thus do not provide reasonable forecasts of individual expected pension benefits, they are a highly useful concept for our purpose since they correspond exactly to the pension rights that have been accrued up to a certain point in time (i.e. they are the “accrued-to-date” value; see section 5.1 below).¹⁴

For early retirement within an age corridor between the age of 62 and 65 there are de- ductions of 5.1% for each year of earlier retirement and supplements of 4.2% for each year after 65 up to the age of 68. Only persons with a record of at least 40 years of insurance can use the pension corridor. Once the initial pension benefits are calculated according to the rules specified above, the ongoing pensions are (typically) adjusted with the rate of inflation. For non-contributory qualifying periods (due to childcare, unemployment, sickness etc.) the pension accounts are credited with specified amounts that are financed from the general government budget.

As stated above, our estimates of pension entitlements are defined as the present value of the entire stream of expected public pension benefits. For the later calculations of these present values (see equation (4) in section 5) it is useful to express the determination of pension benefits in formal terms. The (annual) initial pension benefit PBI_i ≡ PB_i(R_i) for individual i is given by:

PBI_i =κY_iD_i(1−λ×(R_i^s−R_i)) =TC_i×(1−λ×(R^s_i −R_i)), (1) whereκ= 0.0178 is the accrual rate,Y_i is the average lifetime pensionable labor income,

14It should be noted that the pension accounts of the APG were only introduced in the year 2005 for all birth cohorts born 1955 or later. Individuals born before this date remained in the old system while people that entered the labor market in or after 2005 have been covered entirely by the new system.

For individuals born after 1954 that have worked before 2005, however, the original law had stipulated a mixed calculation that contained elements of the new and the old system. This turned out to be rather complicated and in 2013 the old claims were “summarized” in an “initial credit” (“Kontoerstgutschrift”, IC) which was intended to compensate for the discontinuation of the mixed calculation. This initial credit is included in the value of the total credit as reported in the pension account statement. In our datatset—see section 4—we also have information about the precise value of initial credits for all insured individuals.

(18)

D_i stands for the number of contribution (or insurance) years, R_i is the retirement age of individual i, R_i^s is his or her statutory retirement age and λ the annual deduction (supplement) for early (late) retirement (λ = 0.051 for R_i < R^s_i and λ = 0.042 for R_i > R^s_i). Total credits are given by TC_i = κY_iD_i (where for notational simplicity we abstract from the presence of initial credits). Ongoing pensions are adjusted with the rate of inflationπ (for simplicity here assumed to be constant). This means that for ages x > R_i it holds that:

PB_i(x) = PB_i(x−1)(1 +π). (2) A final element one has to take into account when determining the disposable pension incomes is the tax system. It is given by applying the tax rateτ_i(x) (which follows from the income tax schedule) to the gross pension of individual i at age x. This disposable pension incomeP_i(x) can thus be written as:

P_i(x) = (1−τ_i(x))×PB_i(x) = (1−τ_i(x))PB_i(x−1)(1 +π)

= (1−τ_i(x))PBI_i(1 +π)^x−Rⁱ

= (1−τ_i(x))(1 +π)^x−Rⁱ(1−λ×(R^s_i −R_i))×TC_i, (3) where τ_i(x) is the (expected) tax rate at age x and where we use equations (1) and (2) for the substitutions. We use equation (3) later as the basis for the calculation of public pension entitlements.¹⁵

4 Data and methodology

Our main data source for net wealth is the Eurosystem Household Finance and Con- sumption Survey (HFCS) to which we match data from the social security register. Both data sources are briefly described in the following. Details of the data and the statistical matching methodology can be found in Lindner & Sch¨urz (2021).

4.1 HFCS

The HFCS is a comprehensive survey of households’ balance sheets that covers incomes, expenditures as well as real assets, financial assets and debt and thus allows the calcula-

15Importantly, we take the total creditsT C_i²⁰¹⁶ of individual iat the end of 2016 and revalue them up to the year of retirement, i.e. we use hypothetical total credits amounting to: T C_i²⁰¹⁶(1 +g)^Rⁱ^−aⁱ, wherea_i is individuali’s age in 2017 andg is the average revaluation rate.

(19)

tion of net wealth. We use the third wave of the HFCS for Austria that has been carried out between late November 2016 and July 2017. A total of 3,072 households have been interviewed which contain 6,414 persons, 5,476 of which were 16 years old or older and are used for our analysis. In order to deal with the issue of non-responses the HFCS uses multiple imputations. In particular, the dataset provides five imputed values (replicates) for every missing value. In addition, the dataset also contains 1000 replicate weights that can be used to estimate standard errors without having the full sampling information. These weights refer to the household level and we will also use it when presenting weighted results on the person level. The information about persons in the HFCS is somewhat less extensive than on the household level. Nevertheless, it contains a number of questions about occupation, work history, income sources and various pension rights that are valuable for the calculation of pension entitlements. In particular, we have also made use of a set of special variables (e.g. about the pension account statement) that are not part of the harmonized set of core variables present in all participating countries.

4.2 Data from the social security register

The second bulk of data stem from the social security register (SSR). In particular, we managed to obtain a complete snapshot of social security data for the year 2016. These data contain (i) information about the pension account statements of all active individuals born between 1955 and 2001, (ii) information about the pension payments for all retired individuals (except retired civil servants). The data do not contain information about active person that have been born before 1955 since for them the pension account system does not apply (see section 3). The information for the active population (more than 4 million individuals) includes: gender, the age group (in 5-year intervals), the social security institution, the postal code, the initial pension credit, the annual pension credits for the year 2016 and the total pension credits at the end of the year 2016. The information for the retired population (about 650,000 individuals) is similar, only that now the information about the pension account is substituted by data on the monthly gross pension (for December 2016), the pension type (old age, survivor, disability etc.) and the point of time when the pension payments started.

4.3 Statistical matching

In order to amend the information from the wealth survey with information about public pension entitlements the SSR data (the donor) have been matched to the HFCS data (the

(20)

recipient). The matching procedure has been implemented at the person level making use of the following matching variables: age, gender, income, geographical information (postal code), social security institution. Each person implicate in the HFCS data is taken as a separate observation and matched to a specific observation in the SSR data. The matching is based on the random hotdesk procedure and two variants were conducted (that differ in whether income is treated as a categorical or a continuous variable). For details on the matching procedure and the results see Lindner & Sch¨urz (2021).

The first matching procedure (based on income categories) which we also use as our benchmark specification below resulted in an average total pension credit (for individuals) of e11,340 (which is above the unweighted average of the SSR data amounting to e9,800).¹⁶ The median is e9,150 while the highest value is close to e50,000. We can aggregate the individual total credits to the household level and arrive at a (weighted) household mean of e18,500 with a median of e15,500 and a largest value ofe91,500.

5 Calculation of pension entitlements

Once we have the matched values for the individual total pension credits (see the previous section) we can proceed to use this information to transform it into a unique number that can be regarded as a suitable estimation of the present value of public pension entitlements. This process involves several important assumptions as will be discussed in the following.

5.1 “Accrual method” vs. “ongoing concern method”

The first issue in this endeavor is related to the range of expected pension payments that should be included into the present value term of the active population. Should this encompass only entitlements that have been acquired up to the valuation date or should it also cover pension rights that are likely to be gained in the future (after subtracting future pension contributions)? The first approach is typically called the “accrual method”

while the latter approach is referred to as the “ongoing-concern method”. Most papers of the related literature use the first method, although there are a number of important exceptions (see e.g. Sabelhaus & Volz 2020, Catherine et al. 2020). We join the majority

16The difference is likely due to the fact that the SSR contains a larger fraction of people with low or very low incomes in 2016. In fact, every individual who has worked at some time in Austria has a pension account even if he or she did not have any income in 2016. As a consequence the share of individuals with income below the minimum income threshold is higher in the SSR than in the HFCS data.

(21)

of researchers and base our calculations on the accrual method. In our view this approach is in line with the general logic of household surveys (as it excludes future revenues) and as a by-product it also involves less assumptions about the expected working career which necessarily introduce a considerable degree of uncertainty.¹⁷

5.2 The present value formula

In this section we describe how one can use the information on total credits (for active workers) and pension payments (for pensioners) to calculate the present value of public pension entitlements. This present value can be calculated on the basis of the following formula:

PE_i = ^X^ω

x=max(a_i,R_i)

s_i(x) s_i(a_i)

P_i(x)

(1 +δ_i)^x−aⁱ, (4)

where we use the following notation: PE_i stands for the pension entitlements (the present value of expected public pension entitlements) of person i, a_i for his or her age in the year 2016 and R_i for his or her retirement age. For an active (not-retired) person the retirement age lies in the future while for an already retired individual the retirement was an event of the past. The maximum function implements this distinction between active workers (a_i < R_i) and retirees (a_i > R_i) and equation (4) thus applies to both groups. s_i(x) denotes the survival rates, i.e. the probability that person i is still alive at age x ≥a_i.¹⁸ The use of an index i in this expression captures the fact that there exists a strong correlation between specific individual characteristics (like education or income) and mortality. The parameter ω stands for the maximum age (say 110) that is assumed to be the same for all cohorts. δ_i denotes the discount rate that is used today (i.e. in 2016) to discount a pension payment that is delivered in the year 2016 +x−a_i. Note that for a person who has just retired in this year (x=a_i) this first pension benefit is not

17The papers that use the ongoing-concern method often have a different focus. For example, they want to study how the pension system treats different cohorts and if there are changes over time in the degree of sustainability of the system and in government subsidies.

This issue is also related to the appropriate measurement of pensions liabilities (or implicit debt) when one moves from the perspectives of the households (or individuals) to the one of the state (or the pension system). This literature (see Holzmann et al. 2004) makes similar distinctions and works with different definitions of pension liabilities, e.g., “accrued-to-date liabilities”, “projected liabilities of current workers and pensioners” and “open-system liabilities” (ibd., p.12). The concept of “accrued-to-date liabilities”

corresponds to the “accrued assets” that we are going to use as the basis of our calculations.

18Alternatively one could also writes_i(a_i, x) instead of _s^sⁱ^(x)

i(a_i) in equation (4) withs_i(y, x) denoting the survival probability for individualifrom ageyto agex≥y. Note that it holds thats(a_i, x) = _s(0,a^s(0,x)

i)

and thus the two expressions are identical.

(22)

discounted. Otherwise pension payments that lie in the future do not enter the individual valuation at full value as is commonly assumed in these kinds of calculations. There exists a long controversy about the determination and the right choice of discount rates and we come back to this issue below. In order not to jump ahead of this discussion we again allow for the possibility that discount rates differ across individual members of these cohorts. P_i(x) finally stand for the pension income that person iexpects to receive at age x≥max(a_i, R_i) and it has been specified above in equation (3). The termP_i(x) contains the entire effective legislation concerning the pension system. This comprises, on the one hand, the stipulations determining the size of the initial pension and pension adjustments as described above. In addition, the amount of pension income will also depend on whether one uses a gross or net concept where the latter accounts for taxes and social security contributions. Furthermore, one could also try to include the system of survivors’ pension into this framework. Finally, the accurate size of future pension benefits will also depend on the expectation about future changes in the regulations (“pension reforms”) and on the probabilities individuals (or the modelers) assign to the size and the extent of these adjustments.

On the whole, there exist a large number of possible specifications for the various parameters and variables that are necessary to calculate the PE_i. The (somewhat complicated) expression in equation (4) is a distinct reminder of the plethora of assumptions needed to complete these calculations and it also serves as a useful structure to organize later extensions, refinements and discussions.

5.3 Benchmark specification

In the following we list the assumptions that we chose for our benchmark specification and we briefly explain the underlying rationale behind the choices. Details can be found in the appendix while the results of alternative assumptions are discussed in a later section.

• Statutory retirement age. For R_i we assume that every individual will retire at the current statutory retirement age. For men this amounts to the age of 65 while for women it will gradually be raised from 60 to 65 years in the period from 2024 to 2033 (by steps of sixth months). In this case we do not have to take deduction (supplements) for early (late) retirement into account. In section 6.4 we will also look at the case where the retirement age R_i is set equal to individuals’ expected retirement ages.

(23)

• Life expectancy related to household income. Numerous studies covering various countries and time periods have documented that life expectancy of low- income individuals is considerably below the one of high-income individuals. Chetty et al. (2016), for example, report that the life expectancy gap between the richest and the poorest percent of the American population is 14.6 years (for men) and 10.1 years (for women).¹⁹ In order to come up with income-specific mortality rates for Austria we followed a procedure that has also been used by Sabelhaus & Volz (2020) (see their Appendix B). This method starts with the mortality rates for the year 2017 provided by Statistics Austria that are differentiated by gender and age.

In the next step we make an adjustment for differential mortality based on the pattern reported in Chetty et al. (2016) for the US. This adjustment is specified in such a way that theaverage mortality rate corresponds to the officially documented mortality rate for each gender/age group while within each group the mortality rates are allowed to differ with respect to the household income decile with relative mortality rates corresponding to the ones in Chetty et al. (2016). Details of the method are described in appendix A.2.2.²⁰

• Discount rate δ_i = δ = 3%. For the discount rate we assume a rate that is identical for all individuals and given by 3%. This follows the assumptions of and facilitates the comparison with the related literature (B¨onke et al. 2019, Sabelhaus

& Volz 2020). We want to emphasize, however, that this assumption is neither inconsequential nor trivial. In section 6.4 we come back to this issue when we discuss the effect of different choices of the discount rate.

• Net pension benefits: The final element in equation (4) that is necessary to calculate total public pension entitlements PE_i is evidently the stream of future disposable pension incomes P_i(x) (see equation (3)). These entitlements are deter- mined by the specific regulations of the pension system and it is here that one can observe a huge amount of cross-country diversity. For the calculations of the initial pension benefit we need an assumption on the real growth rate (chosen asg = 1.3%,

19Data for Germany can be found in von Gaudecker & Scholz (2007) and Breyer & Hupfeld (2009).

20One might argue that the pattern of relative mortality differences for the US is not applicable to Austria. The available data, however, do not confirm this conjecture. In particular, there exists a study for Norway—a country with an arguably similar demographic structure as Austria (life expectancy of 82.5 vs. 81.8) —that finds very similar results to the ones reported in Chetty et al. (2016) for the US:

“The difference in life expectancy between the richest and poorest 1% was 8.4 years for women and 13.8 years for men. The differences widened between 2005 and 2015 and were comparable to those in the United States” (Kinge et al. 2019).

(24)

a value that is in line with the assumptions made in European Commission (2021)) and an assumption on tax treatment. We use a net concept since in Austria contributions to the public pension system are exempt from taxation while the pensions payments are treated like earned income and are subject to income tax (for details see appendix A.3).²¹ In the benchmark specification we do not take survivor and minimum pensions into account. Also we assume that all individuals will meet the conditions concerning the minimum insurance years that are necessary to be eligible for a pension.

In section 6.4 we discuss how the results of the benchmark specification are affected if the main assumptions are changed.

6 Results

In the following we will first present the estimates for the main aggregates (net wealth, pension entitlements and augmented wealth) in section 6.1 before we show various socio- economic breakdowns of the aggregate measures (section 6.2) and the implications for distribution and wealth inequality (section 6.3). In section 6.4 we show how the benchmark results change for alternative specifications and in section 6.5 we discuss some limitations of our analysis.

6.1 Aggregates and wealth composition

In this section we present estimates for households’ net wealth, total pension entitlements and augmented wealth. The results for net wealth correspond to the figures reported in Fessler et al. (2019) and are discussed there in detail. In Table 2 we only repeat some important measures for the sake of comparison with the novel results.²² In particular, we show the unconditional means and medians, the fraction of households that holds a specific wealth component (last column) and finally the mean and median for this subsample of households (i.e. the conditional values). For net wealth the conditional and unconditional measures coincide since every household holds some form of wealth (or liability).

21On different models of how to tax pension contributions and payments see Genser & Holzmann (2021).

22Following the standard practice in the analysis of HFCS data the computations are based on a bootstrap procedure using the five multiply imputed datasets.

(25)

The mean of household net wealth is around e250,000 which is considerably larger than the median of around e83,000. This is an indication of the unequal distribution of net wealth across household. We come back to the issue of the wealth distribution below in section 6.3. In Table 2 we also report aggregate measures for the most important subcomponents of net wealth which is defined as the sum of real assets and financial assets minus total debt. We have further subdivided real assets into the main residence, investment in self-employed business and all other real assets. The information about the ownership of the main residence is interesting since it represents for many households the most important category of wealth. In particular, almost 46% of households are owners of their main residence with a (conditional) mean value of e289,000 and a median of e250,000. For the subgroup of owners the conditional mean of net wealth is arounde470,000 and thus the value of the main residence is on average 60% of their total wealth. Investment in unincorporated enterprises, on the other hand, is only observed for a minority of households (7%) for whom, however, the conditional mean is fairly high (e662,000).

Financial assets and other (real) assets (like vehicles, valuables and other real estate property) are less important for household net wealth with means of arounde39,000 and e51,000, respectively. The participation rates in these categories are, however, high with 99.7% and 83.3%. Finally, the conditional mean of total (collateralized and uncollater- alized) debt is around e57,000 where it is important to note that in Austria more than two-thirds of households do not have any debt (which is arguably the mirror image of the comparably low rate of home ownership).

We turn now to the estimates for household (public) pension entitlements. For this we added up the measures of the individual present values of pension entitlements that have been calculated from equation (4) under the assumptions specified above. We find that almost all household (around 99%) either receive or can expect to receive a pension²³ and that these (intangible) pension entitlements are large when compared to the tangible measure of net wealth. In particular, the median is arounde200,000 which is more than twice the median of net wealth while the mean is about e245,000 which is only slightly below the mean value for net wealth.

One can add up net wealth and pension entitlements to arrive at a measure of “augmented wealth”.²⁴ As discussed above, this is not innocuous since it adds two different

23The households without pension entitlements are mostly households that consist of young adults who have not yet contributed to the public pension system.

24The concept of “augmented wealth” is typically defined as also including the present value of occu-

(26)

(strictly spoken incommensurable) categories of wealth but it is a common practice in the related literature. The median comes out as around e330,000 which is four times larger than the median of net wealth. The mean, on the other hand, is calculated as e495,000 which is thus about twice as large as the mean for net wealth.

Table 2: Aggregates and wealth composition

Wealth aggregate Mean Median Cond. Mean Cond. Median Participation

(ine) (ine) (ine) (ine) (in %)

Net wealth 250,272 82,681 250,272 82,681 100.00

(23,547) (3,301) (23,547) (3,301) (0)

Financial assets 38,637 15,460 38,738 15,539 99.74

(1,927) (735) (1,930) (736) (0.1)

Main residence 132,825 0 289,112 250,000 45.94

(2,851) (0) (5,772) (0) (4967)

Investment in self-employed business (incl. farms) 46,284 0 661,534 108,133 7.00

(19,960) (0) (279,724) (37,446) (0.6)

Other (real) assets 51,292 7,900 61,570 10,000 83.31

(5,940) (346) (6,939) () (0.7)

Total debt 18,766 0 57,328 17,140 32.73

(1,395) (0) (3,768) (1,869) (1.0)

Public pension entitlements 245,051 197,232 247,605 199,530 98.97

(4,169) (5,046) (4,026) (4,724) (0.2)

Augmented wealth 495,324 327,456 495,324 327,456 100.00

(24,762) (6,187) (24,762) (6,187) (0)

Note: This table shows statistics for various aggregates of households’ wealth and pension entitlements.

The last column shows the percentage of households that have non-empty (but not necessarily positive) entries in the respective category. The conditional means and medians are the aggregate values for all household with non-empty observations. All statistics are based on imputed values. Bootstrapped standard errors using 1000 replica weights are shown in brackets.

6.2 The influence of socio-economic characteristics on wealth and pension entitlements

In this section we report the influence of various socio-economic characteristics on households’ pension entitlements. In the first subsection 6.2.1 we approach this topic by presenting breakdowns with respect to various socio-economic indicators and we compare the results to the breakdowns with respect to net wealth. In subsection 6.2.2 we use a (multivariate) regression analysis to study the factors that have the largest impact on these entitlements.

pational pensions. Due to the lack of reliable data we leave out occupational pensions in our benchmark estimation. Since the second pillar only plays a minor role in Austria this omission is likely to have only negligible effects on the results as is confirmed in a robustness exercise in section 6.4.

Pension Entitlements and Net Wealth in

WORKING PAPER 238

Pension Entitlements and Net Wealth in

Austria

Pension Entitlements and Net Wealth in Austria ∗

Markus Knell

and Reinhard Koman

Oesterreichische Nationalbank February 2022

Non-Technical Summary

Contents

1 Introduction

2 Related literature

3 The Austrian pension system

4 Data and methodology

4.1 HFCS

4.2 Data from the social security register

4.3 Statistical matching

5 Calculation of pension entitlements

5.1 “Accrual method” vs. “ongoing concern method”

5.2 The present value formula

5.3 Benchmark specification

6 Results

6.1 Aggregates and wealth composition

6.2 The influence of socio-economic characteristics on wealth and pension entitlements

Pension Entitlements and Net Wealth in Austria ^∗